add login and make it secure,
use the db we already haveOpen source · Remote first · v1.0.0
Your powerful desktop, available from your portable IDEAI Prompt Optimizer
Select the files that matter in the current project. Add Web Search or Deep Research when needed. Then use the strongest machine you own from any IDE terminal.
Independent MIT-licensed project; not affiliated with OpenAI, Google or Anthropic.
The control point most prompt tools skip
It does not optimize blind.
Point it at the current project, choose the exact files, and keep the rest out. The output is grounded in the architecture you actually have—not a guessed stack.
- Open current project
- Choose exact files
- Add task and research
- Review ready prompt
- Run it anywhere
Rough in. Structured out.
Turn a vague task into an executable specification.
Most failed AI coding sessions begin before the model answers: the request hides constraints, leaves success undefined and invites the executor to invent product decisions. AI Prompt Optimizer compiles the selected context and intent into a prompt you can inspect before execution.
- Preserved requirements and explicit constraints
- Bounded assumptions and unresolved questions
- Implementation and failure-handling plan
- Security checks and testable acceptance criteria
One compiler, portable output
Optimize prompts for the AI tool you already use.
Clearer implementation prompts
Compile the task first, review the assumptions, then paste the vendor-neutral result into your selected OpenAI workflow.
Structured Gemini prompts
Use the same compiled specification with Gemini without coupling AI Prompt Optimizer output to a proprietary prompt template.
Better coding-agent context
Make scope, constraints, verification and completion criteria explicit before an agent edits the repository.
No downstream lock-in
The artifact is readable plain text, so you can compare executors or keep the prompt with the engineering record.
Desktop power, laptop mobility
Keep the workstation active. Work from anywhere.
The home workstation runs AI Prompt Optimizer, the supervised llama-server and the local model. Your MacBook or lighter laptop runs only the interactive client from the IDE terminal and keeps the active project workspace.
- At home: an always-on Windows or Ubuntu machine provides GPU inference.
- On the move: the portable IDE connects through an authenticated WebSocket.
- At execution: the optimized text remains portable across coding agents.
Use a private overlay network or VPN; direct public port forwarding is not recommended. Remote exposure is opt-in and HMAC-authenticated, while private networking and TLS remain operator responsibilities.
Published evaluation
Measured improvement, bounded claim.
0.333 → 0.792
Eight stratified local cases compared the raw request with the compiled output.
0.725 → 0.975
Explicit-requirement precision remained 1.000; recall changed from 1.000 to 0.917.
141 automated tests
53 Bun, 79 pytest and 9 integration tests cover provider, protocol, fallback and confinement paths.
No universal superiority claim
The reference is a small lexical evaluation on one workstation, not human judgment or downstream task-completion proof.
AI prompt optimizer FAQ
Direct answers before you run it.
What is an AI prompt optimizer?
It converts a rough request and selected project context into a clearer, more testable prompt before another AI executes it. AI Prompt Optimizer compiles developer intent into requirements, constraints, assumptions, steps and acceptance criteria.
Does AI Prompt Optimizer work with ChatGPT, Gemini, Claude Code and Codex?
Yes as an output workflow. AI Prompt Optimizer produces portable text that can be reviewed and pasted into those tools and other coding agents. They are downstream targets, not native plugins or official integrations.
Is AI Prompt Optimizer an official OpenAI prompt optimizer?
No. AI Prompt Optimizer is independent and MIT licensed. It can use an operator-configured OpenAI-compatible endpoint, but it is not affiliated with or endorsed by OpenAI.
Can it run on an always-on home GPU workstation?
Yes. Keep the server and model on a Windows or Ubuntu workstation at home, then run the client from a MacBook or lighter laptop IDE terminal over a protected private network.
Does it have a native IDE extension?
Not currently. The supported workflow runs the Bun client from the IDE integrated terminal and connects it to the remote AI Prompt Optimizer server.
Start without a GPU
Test the complete workflow in three minutes.
The deterministic mock path uses the same application boundary as configured inference, requires no credentials and writes the optimized prompt to a local artifact.